| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <20251117132537.227116-1-lihongbo22@huawei.com>
Date: Mon, 17 Nov 2025 13:25:27 +0000
From: Hongbo Li <lihongbo22@...wei.com>
To: <hsiangkao@...ux.alibaba.com>, <chao@...nel.org>, <brauner@...nel.org>,
<djwong@...nel.org>, <amir73il@...il.com>, <joannelkoong@...il.com>
CC: <lihongbo22@...wei.com>, <linux-fsdevel@...r.kernel.org>,
<linux-erofs@...ts.ozlabs.org>, <linux-kernel@...r.kernel.org>
Subject: [PATCH v9 00/10] erofs: inode page cache share feature
Enabling page cahe sharing in container scenarios has become increasingly
crucial, as it can significantly reduce memory usage. In previous efforts,
Hongzhen has done substantial work to push this feature into the EROFS
mainline. Due to other commitments, he hasn't been able to continue his
work recently, and I'm very pleased to build upon his work and continue
to refine this implementation.
This patch series is based on Hongzhen's original EROFS shared pagecache
implementation which was posted about half a year ago:
https://lore.kernel.org/all/20250301145002.2420830-1-hongzhen@linux.alibaba.com/T/#u
In addition to the forward-port, I have also fixed several bugs, resolved
some prerequisite dependencies and performed some minor cleanup.
(A recap of Hongzhen's original cover letter is below, edited slightly
for this serise:)
Background
==============
Currently, reading files with different paths (or names) but the same
content can consume multiple copies of the page cache, even if the
content of these caches is identical. For example, reading identical
files (e.g., *.so files) from two different minor versions of container
images can result in multiple copies of the same page cache, since
different containers have different mount points. Therefore, sharing
the page cache for files with the same content can save memory.
Proposal
==============
1. determining file identity
----------------------------
First, a way needs to be found to check whether the content of two files
is the same. Here, the xattr values associated with the file
fingerprints are assessed for consistency. When creating the EROFS
image, users can specify the name of the xattr for file fingerprints,
and the corresponding name will be stored in the packfile. The on-disk
`ishare_key_start` indicates the index of the xattr name within the
prefix xattrs:
```
struct erofs_super_block {
__u8 xattr_filter_reserved; /* reserved for xattr name filter */
- __u8 reserved[3];
+ __u8 ishare_key_start; /* start of ishare key */
+ __u8 reserved[2];
};
```
For example, users can specify the first long prefix as the name for the
file fingerprint as follows:
```
mkfs.erofs --ishare_key=trusted.erofs.fingerprint erofs.img ./dir
```
In this way, `trusted.erofs.fingerprint` serves as the name of the xattr
for the file fingerprint. The relevant patch for erofs-utils has been posted
in:
https://lore.kernel.org/all/20251114092845.207368-1-lihongbo22@huawei.com/
At the same time, for security reasons, this patch series only shares
files within the same domain, which is achieved by adding
"-o domain_id=xxxx" during the mounting process:
```
mount -t erofs -o domain_id=trusted.erofs.fingerprint erofs.img /mnt
```
If no domain ID is specified, it will fall back to the non-page cache
sharing mode.
2. Implementation
==================
2.1. file open & close
----------------------
When the file is opened, the ->private_data field of file A or file B is
set to point to an internal deduplicated file. When the actual read
occurs, the page cache of this deduplicated file will be accessed.
When the file is opened, if the corresponding erofs inode is newly
created, then perform the following actions:
1. add the erofs inode to the backing list of the deduplicated inode;
2. increase the reference count of the deduplicated inode.
The purpose of step 1 above is to ensure that when a real I/O operation
occurs, the deduplicated inode can locate one of the disk devices
(as the deduplicated inode itself is not bound to a specific device).
Step 2 is for managing the lifecycle of the deduplicated inode.
When the erofs inode is destroyed, the opposite actions mentioned above
will be taken.
2.2. file reading
-----------------
Assuming the deduplication inode's page cache is PGCache_dedup, there
are two possible scenarios when reading a file:
1) the content being read is already present in PGCache_dedup;
2) the content being read is not present in PGCache_dedup.
In the second scenario, it involves the iomap operation to read from the
disk.
2.2.1. reading existing data in PGCache_dedup
-------------------------------------------
In this case, the overall read flowchart is as follows (take ksys_read()
for example):
ksys_read
│
│
▼
...
│
│
▼
erofs_ishare_file_read_iter (switch to backing deduplicated file)
│
│
▼
read PGCache_dedup & return
At this point, the content in PGCache_dedup will be read directly and
returned.
2.2.2 reading non-existent content in PGCache_dedup
---------------------------------------------------
In this case, disk I/O operations will be involved. Taking the reading
of an uncompressed file as an example, here is the reading process:
ksys_read
│
│
▼
...
│
│
▼
erofs_ishare_file_read_iter (switch to backing deduplicated file)
│
│
▼
... (allocate pages)
│
│
▼
erofs_read_folio/erofs_readahead
│
│
▼
... (iomap)
│
│
▼
erofs_iomap_begin
│
│
▼
...
Iomap and the layers below will involve disk I/O operations. As
described in 2.1, the deduplicated inode itself is not bound to a
specific device. The deduplicated inode will select an erofs inode from
the backing list (by default, the first one) to complete the
corresponding iomap operation.
2.3. release page cache
-----------------------
Similar to overlayfs, when dropping the page cache via .fadvise, erofs
locates the deduplicated file and applies vfs_fadvise to that specific
file.
Effect
==================
I conducted experiments on two aspects across two different minor
versions of container images:
1. reading all files in two different minor versions of container images
2. run workloads or use the default entrypoint within the containers^[1]
Below is the memory usage for reading all files in two different minor
versions of container images:
+-------------------+------------------+-------------+---------------+
| Image | Page Cache Share | Memory (MB) | Memory |
| | | | Reduction (%) |
+-------------------+------------------+-------------+---------------+
| | No | 241 | - |
| redis +------------------+-------------+---------------+
| 7.2.4 & 7.2.5 | Yes | 163 | 33% |
+-------------------+------------------+-------------+---------------+
| | No | 872 | - |
| postgres +------------------+-------------+---------------+
| 16.1 & 16.2 | Yes | 630 | 28% |
+-------------------+------------------+-------------+---------------+
| | No | 2771 | - |
| tensorflow +------------------+-------------+---------------+
| 2.11.0 & 2.11.1 | Yes | 2340 | 16% |
+-------------------+------------------+-------------+---------------+
| | No | 926 | - |
| mysql +------------------+-------------+---------------+
| 8.0.11 & 8.0.12 | Yes | 735 | 21% |
+-------------------+------------------+-------------+---------------+
| | No | 390 | - |
| nginx +------------------+-------------+---------------+
| 7.2.4 & 7.2.5 | Yes | 219 | 44% |
+-------------------+------------------+-------------+---------------+
| tomcat | No | 924 | - |
| 10.1.25 & 10.1.26 +------------------+-------------+---------------+
| | Yes | 474 | 49% |
+-------------------+------------------+-------------+---------------+
Additionally, the table below shows the runtime memory usage of the
container:
+-------------------+------------------+-------------+---------------+
| Image | Page Cache Share | Memory (MB) | Memory |
| | | | Reduction (%) |
+-------------------+------------------+-------------+---------------+
| | No | 34.9 | - |
| redis +------------------+-------------+---------------+
| 7.2.4 & 7.2.5 | Yes | 33.6 | 4% |
+-------------------+------------------+-------------+---------------+
| | No | 149.1 | - |
| postgres +------------------+-------------+---------------+
| 16.1 & 16.2 | Yes | 95 | 37% |
+-------------------+------------------+-------------+---------------+
| | No | 1027.9 | - |
| tensorflow +------------------+-------------+---------------+
| 2.11.0 & 2.11.1 | Yes | 934.3 | 10% |
+-------------------+------------------+-------------+---------------+
| | No | 155.0 | - |
| mysql +------------------+-------------+---------------+
| 8.0.11 & 8.0.12 | Yes | 139.1 | 11% |
+-------------------+------------------+-------------+---------------+
| | No | 25.4 | - |
| nginx +------------------+-------------+---------------+
| 7.2.4 & 7.2.5 | Yes | 18.8 | 26% |
+-------------------+------------------+-------------+---------------+
| tomcat | No | 186 | - |
| 10.1.25 & 10.1.26 +------------------+-------------+---------------+
| | Yes | 99 | 47% |
+-------------------+------------------+-------------+---------------+
It can be observed that when reading all the files in the image, the
reduced memory usage varies from 16% to 49%, depending on the specific
image. Additionally, the container's runtime memory usage reduction
ranges from 4% to 47%.
[1] Below are the workload for these images:
- redis: redis-benchmark
- postgres: sysbench
- tensorflow: app.py of tensorflow.python.platform
- mysql: sysbench
- nginx: wrk
- tomcat: default entrypoint
Changes from v8:
- add review-by in patch 1 and patch 10.
- do some clean up in patch 2 and patch 4,6,9 as suggested by Xiang.
- add new patch 3 to export alloc_empty_backing_file.
- patch 5 only use xattr prefix id to record the ishare info, changed
config to EROFS_FS_PAGE_CACHE_SHARE and make it compatible.
- patch 7 use backing file helpers to alloc file when ishare file is
opened as suggested by Xiang.
- patch 8 remove erofs_read_{begin,end} as suggested by Xiang.
v8: https://lore.kernel.org/all/20251114095516.207555-1-lihongbo22@huawei.com/
v7: https://lore.kernel.org/all/20251021104815.70662-1-lihongbo22@huawei.com/
v6: https://lore.kernel.org/all/20250301145002.2420830-1-hongzhen@linux.alibaba.com/T/#u
v5: https://lore.kernel.org/all/20250105151208.3797385-1-hongzhen@linux.alibaba.com/
v4: https://lore.kernel.org/all/20240902110620.2202586-1-hongzhen@linux.alibaba.com/
v3: https://lore.kernel.org/all/20240828111959.3677011-1-hongzhen@linux.alibaba.com/
v2: https://lore.kernel.org/all/20240731080704.678259-1-hongzhen@linux.alibaba.com/
v1: https://lore.kernel.org/all/20240722065355.1396365-1-hongzhen@linux.alibaba.com/
Diffstat:
fs/erofs/Kconfig | 9 ++
fs/erofs/Makefile | 1 +
fs/erofs/data.c | 89 +++++++++----
fs/erofs/erofs_fs.h | 6 +-
fs/erofs/fscache.c | 13 --
fs/erofs/inode.c | 4 +
fs/erofs/internal.h | 50 ++++++++
fs/erofs/ishare.c | 283 +++++++++++++++++++++++++++++++++++++++++
fs/erofs/super.c | 53 +++++++-
fs/erofs/xattr.c | 15 +++
fs/erofs/zdata.c | 42 ++++--
fs/file_table.c | 1 +
fs/fuse/file.c | 4 +-
fs/iomap/buffered-io.c | 6 +-
include/linux/iomap.h | 8 +-
15 files changed, 524 insertions(+), 60 deletions(-)
create mode 100644 fs/erofs/ishare.c
--
2.22.0
Powered by blists - more mailing lists